Plan analysis: drop three rules PerformanceStudio removed, and show its CPU:Elapsed ratio (#4537)

[erikdarlingdata]

Fixes #4537
Part of #4511

Why

PerformanceStudio removed three plan-analysis rules after its rule-set review (erikdarlingdata/PerformanceStudio#215). PerformanceMonitor still ran them. The two analyzers are meant to be equivalent, so this removes them here too.

• Rule 21, CTE Multiple References, was removed in erikdarlingdata/PerformanceStudio@aabbaa2 (Strip AI-drafted wait stat content + legacy marker + drop Rule 21 (#215 D3) PerformanceStudio#266). An actual plan's runtime stats show where the time went, so a warning based on a text pattern for CTE reuse is a guess.
• Rule 25, Ineffective Parallelism, and rule 31, Parallel Wait Bottleneck, were removed in erikdarlingdata/PerformanceStudio@28d4c74. These warnings guessed at causes that the wait stats already show. In their place, PerformanceStudio's runtime summary shows a CPU:Elapsed ratio, which the user can read against the DOP.

What changes

• PlanAnalyzer:
  • removes rules 21, 25 and 31;
  • removes the helpers only they used (rule 21's CTE detector, and the wait-stats advice helpers that rules 25 and 31 called).
• BenefitScorer: removes the rule 25 and rule 31 scoring cases.
• The wait-stat half of erikdarlingdata/PerformanceStudio@aabbaa2 is already matched: PerformanceMonitor reads curated descriptions from Resources/WaitStats.json (WaitStatsConfig.Description), and EmitWaitStatWarnings sets no ActionableFix. That is PerformanceStudio's current shape.
• The plan viewer's runtime summary gains the CPU:Elapsed row, as PerformanceStudio's does:
  • it sits directly after Elapsed;
  • it shows only when elapsed time is above zero;
  • it takes external-wait time out of CPU first (BenefitScorer.IsExternalWait), as PerformanceStudio does.
  • The calculation is PlanDisplayText.CpuElapsedRatio, so it can be tested without a WPF host.
• Not ported here:
  • PerformanceStudio adds the CPU:Elapsed row to its web runtime card and HTML export, and PerformanceMonitor has neither.
  • The legacy-rule marker from erikdarlingdata/PerformanceStudio@aabbaa2 (IsLegacy, shown as " [legacy]") is its own change, Port PerformanceStudio's legacy-rule marker: IsLegacy on warnings from pre-framework rules, shown as [legacy] #4566.

Test plan

Darling/Darling.Tests/PlanSync4537Tests.cs has seven tests:

• Three rule-removal tests, one per removed rule. Each builds the plan shape that fired the rule on dev:

  • a CTE referenced twice;
  • a parallel plan with elapsed time well above CPU;
  • a parallel plan with CPU close to elapsed but far below its DOP potential.

  Each asserts the finding is gone. All three were RED on dev, in a detached worktree, and GREEN here. Mutation: restoring the rule 25/31 block turned both of their tests RED.

• Four CpuElapsedRatio tests:

  • CPU 800 ms / elapsed 400 ms with no waits → 2.00;
  • the same with a 300 ms external wait → 1.25;
  • elapsed 0 → no row;
  • an external wait larger than CPU → 0.

  Mutation: removing the external-wait subtraction turned two of them RED.

• Run: PlanSync4537Tests 7/7. The full plan-analysis class list: 238 passed, 0 failed, and 2 skipped (live PostgreSQL, run in CI).

• Builds: Darling.Tests, PerformanceMonitor.Ui, Lite, Lite.Tests and Dashboard.Tests (Release, EnableWindowsTargeting) all have 0 warnings.

CHANGELOG

SECTION: Changed
ENTRY: - Plan analysis drops three rules that PerformanceStudio removed ([#4565]) - CTE Multiple References, Ineffective Parallelism and Parallel Wait Bottleneck are no longer reported. They guessed at what the runtime stats and wait stats already show. The plan viewer's runtime summary shows a CPU:Elapsed ratio instead, with external-wait time left out of CPU, as PerformanceStudio does.
REF: [#4565]: #4565